On June 9, 2026, Anthropic released Claude Fable 5, its first "Mythos-class" model available for general use, alongside Claude Mythos 5, a restricted version of the same model. The two share identical capabilities and differ only in their safeguards: Fable 5 includes safety classifiers that can decline requests about cybersecurity exploitation, dangerous biology and chemistry, or model distillation, while Mythos 5 lifts those restrictions for vetted cyber defenders in Anthropic's Project Glasswing. For developers, Fable 5 introduced a new integration pattern: a frontier model that sometimes declines, and an API designed around retrying those requests on another model.

Key Facts#

  • Release date: June 9, 2026. API model IDs claude-fable-5 (generally available) and claude-mythos-5 (Project Glasswing participants only).
  • Pricing per million tokens: $10 input and $50 output for both models, which Anthropic said is less than half the price of Claude Mythos Preview.
  • Context and output: a one-million-token context window by default and up to 128,000 output tokens per request.
  • Safeguards: Fable 5's classifiers cover cybersecurity, biology and chemistry, and distillation. Anthropic said declined requests are handled by Claude Opus 4.8 in fewer than 5% of sessions on average.
  • Refusal behavior: the Messages API returns stop_reason: "refusal" with HTTP 200. Requests refused before any output is generated are not billed.
  • Thinking: adaptive thinking is always on and cannot be disabled; the raw chain of thought is never returned.
  • Data retention: 30-day retention, and no zero-data-retention option unless Anthropic expressly authorizes it.
  • Availability: the Claude API and apps, Amazon Bedrock, Claude Platform on AWS, Google Cloud and Microsoft Foundry.

What Happened#

Anthropic described Fable 5 as state of the art across nearly all the benchmarks it tested, with particular strength in software engineering, knowledge work, vision, scientific research and long-running autonomous tasks. It cited results such as the top score among frontier models on Cognition's FrontierCode evaluation at medium effort and the highest score on Hebbia's finance benchmark. On science, Anthropic said scientists preferred Mythos-generated hypotheses about 80% of the time in blind comparisons for molecular biology. Launch partners included Stripe, which described compressing months of engineering work into days, as well as GitHub and Cursor.

The safety design was the real news. Instead of training one model to refuse more, Anthropic shipped one model twice. Fable 5 routes flagged requests to classifiers and, when needed, to the less capable Opus 4.8, while Mythos 5 keeps the full capability for approved defenders. Anthropic said external red-teaming produced no universal jailbreaks in more than 1,000 hours, and that the model complied with zero harmful single-turn requests across 30 jailbreak techniques tested.

The API changes followed from that design. Anthropic's developer documentation lists three ways to handle a refusal: a server-side fallbacks parameter, in beta on the Claude API, that retries automatically on recommended models; client-side SDK middleware; or a manual retry. A "fallback credit" refunds the prompt-cache cost of switching models, so developers do not pay twice.

Background#

The Mythos name first became public in March 2026, and on April 7, 2026, Anthropic launched Project Glasswing, giving a small group of organizations access to Claude Mythos Preview to find and fix software vulnerabilities. That limited release established the idea that some models are too capable in sensitive areas for unrestricted access.

Fable 5 was Anthropic's answer to broad access. CNBC characterized it as a Mythos-like model released to the public. It arrived less than two weeks after Claude Opus 4.8, which stayed at $5 and $25 per million tokens and became the fallback model for Fable 5's refusals. With Fable 5 at twice Opus pricing, Anthropic's lineup now had a clear premium tier above Opus for the first time.

Why It Matters for Developers#

Refusals are now a normal response type, not an exception. Because the API returns HTTP 200 with a refusal stop reason, code that only checks for HTTP errors will silently show users an empty answer. In .NET, the official Anthropic SDK's IChatClient implementation maps a refusal to ChatFinishReason.ContentFilter, which makes a client-side fallback straightforward with Microsoft.Extensions.AI:

C#
using Anthropic;
using Microsoft.Extensions.AI;

// AnthropicClient reads ANTHROPIC_API_KEY from the environment.
AnthropicClient anthropic = new();
IChatClient fable = anthropic.AsIChatClient("claude-fable-5");
IChatClient opus = anthropic.AsIChatClient("claude-opus-4-8");

string prompt = "Summarize the security findings in this audit report: ...";
var response = await fable.GetResponseAsync(prompt);

if (response.FinishReason == ChatFinishReason.ContentFilter)
{
    // Fable 5 declined; retry on the model Anthropic uses for its own fallback.
    response = await opus.GetResponseAsync(prompt);
}

Console.WriteLine(response.Text);

Log every refusal with enough context to review it, and track the refusal rate per feature; a spike usually means prompts are drifting into restricted territory. Our guide to responsible AI and LLM security covers monitoring and user messaging.

Compliance teams should also note the 30-day data retention. Many enterprises rely on zero-data-retention agreements, and Fable 5 is excluded from them by default, which can block adoption for regulated workloads until reviewed.

Finally, cost. At $10 and $50 per million tokens, Fable 5 costs twice as much as Opus 4.8, and adaptive thinking cannot be switched off. Use the effort parameter to control depth, and route only the hardest tasks to Fable 5.

What's Next#

Fable 5's first weeks were turbulent. On June 12, 2026, US export controls forced Anthropic to suspend Fable 5 and Mythos 5 for all customers; the controls were lifted on June 30, and access returned on July 1 with an improved classifier that Anthropic said blocks the reported bypass technique in over 99% of cases. Our policy coverage of the export-control suspension has the details.

Anthropic then released Claude Opus 5 on July 24, Claude Fable 5.1 and Mythos 5.1 on September 1 at the same $10 and $50 pricing, and Claude Opus 5.5 on September 22. Fable 5 remains active, with a tentative retirement date of no sooner than June 9, 2027.

Sources#